There has been an increasing interest in moving computation closer to storage in recent years due to significant improvements in memory technology. FPGAs were proven to be an exciting candidate for accelerating database workloads since they provide an energy-efficient, reconfigurable and high-performance computation platform. Therefore, FPGAs are widely used as attached accelerators on data-centric applications.

Database operations usually run on large volumes of data, which creates an I/O bottleneck when processing them on CPUs. Therefore, recently, researchers have been investigating query pushdown techniques during a database load operation. A well-known columnar storage format, Apache Parquet, provides an efficient way to store a database. In addition, current big data processing engines provide functionalities for pushing filter operation down to the parquet reading stage.

This study explores the boundaries of pushing down analytic queries to the parquet reader stage by using FPGAs. An extended roofline analysis is performed on a proof-of-concept hardware design. The analysis shows that peak performance is achieved via a storage-attached accelerator once a high bandwidth interface is introduced. Furthermore, using multiple FPGAs with flash storage while interfacing them with OpenCAPI or PCI switch enables higher performance for aggregation since aggregation is shown to be I/O bound.

The thesis introduces Apache Spark integration of the proof-of-concept query pushdown for parquet reading operations. Apache Spark implements several layers of parallelism to achieve higher speed-ups. However, the concurrency and parallelism for a single FPGA instance for multi-threaded Apache Spark applications requires synchronization on a constrained resource represented by a single FPGA. Therefore, this work suggests a way to achieve synchronization with a single FPGA instance.

The present work shows that for a single Spark thread, a maximum end-to-end application speed-up of 3.88x and a kernel speed-up of 7.24x are achieved. As a result, the throughput of TPC-H Query 6 can be increased up to 3.8 GB/s. Furthermore, FPGA can perform better than CPU until Spark is configured to run on 7 CPU threads. Then, for the scaled-up multi-threaded Spark application with six CPU threads, the FPGA can achieve 1.13x end-to-end application speed-up and a kernel speed-up of 13.19x.

Big data applications are becoming more commonplace due to an abundance of digital data and increasingly powerful hardware. One of these classes of hardware devices are FPGAs, which are being used today in various ways such as data centers and embedded systems. High performance, power efficiency, and reprogrammability are the primary reasons behind their wide use. Another trend over the previous years has been to use distributed data processing frameworks such as Apache Spark to improve the performance of big data applications. Traditionally, such frameworks are deployed on commodity hardware to save costs. This approach is fairly popular, with organizations often having on-premise compute clusters or using a cloud provider to access a managed cluster. This project attempts to combine the above-mentioned worlds - FPGAs and dis- tributed data processing. We have designed an architecture that allows us to use FP- GAs as end-devices in a compute cluster to perform the actual computation instead of CPUs. This architecture is designed by composing together several open source technologies and allows us to interact with an FPGA cluster using Python. Using a high-level programming language such as Python makes this system easy to use for software developers and data scientists, and also abstracts away the internal commu- nication within the cluster. We have built prototypes based on this architecture for 3 hardware platforms (FPGA families) and 3 specific applications to demonstrate general applicability. We have observed noticeable performance gains in these applications by scaling up the FPGA cluster.

Through new digital business models, the importance of big data analytics continuously grows. Initially, data analytics clusters were mainly bounded by the throughput of network links and the performance of I/O operations. With current hardware development, this has changed, and often the performance of CPUs and memory access became the new limiting factor. Heterogeneous computing systems, consisting of CPUs and other computing hardware, such as GPUs and FPGAs, try to overcome this by offloading the computational work to the best suitable hardware.

Accelerating the computation by offloading work to special computing hardware often requires specialized knowledge and extensive effort. In contrast, Apache Spark became one of the most used data analytics tools, among other reasons, because of its user-friendly API. Notably, the component Spark SQL allows defining declarative queries without having to write any code. The present work investigates to reduce this gap and elaborates on how Spark SQL's internal information can be used to offload computations without the user having to configure Spark further.

Thereby, the present work uses the Apache Arrow in-memory format to exchange data efficiently between different accelerators. It evaluates Spark SQL's extensibility for providing custom acceleration and its new columnar processing function, including the compatibility with the Apache Arrow format. Furthermore, the present work demonstrates the technical feasibility of such an acceleration by providing a Proof-of-Concept implementation, which integrates Spark with tools from the Arrow ecosystem, such as Gandiva and Fletcher. Gandiva uses modern CPUs' SIMD capabilities to accelerate computations, and Fletcher allows the execution of FPGA-accelerated computations. Finally, the present work demonstrates that already for simple computations integrating these accelerators led to significant performance improvements. With Gandiva the computation became 1.27 times faster and with Fletcher even up-to 13 times.

An increase in the performance of mobile devices has started a revolution in deploying artificial intelligence (AI) algorithms on mobile and embedded systems. In addition, fueled by the need for privacy-aware insights into data, we see a strong push towards federated machine learning, where data is stored locally and not shared with a central server. By allowing data to stay on client devices and do training locally, we work towards a more privacy-friendly future. Furthermore, utilizing federated machine learning enables machine learning in data-constrained environments where bandwidth is not sufficient to upload the entire dataset. In this thesis, we look at the recent trend into less complex machine learning models. These models optimize resource usage while reducing accuracy loss. We investigate how these simpler models hold up within a federated setting. We also look into the developments of AI frameworks and their capabilities for mobile platforms. Based on these findings, we propose that model-hyper-parameter optimization is possible to maximize accuracy for smaller networks during federated learning. We show that it is possible to reduce the accuracy loss from 15% to only 0.04%. We then demonstrate what a mobile implementation looks like and the performance we see from an iPhone X. We show that an iPhone implementation takes less than 2x the amount of a regular laptop implementation. Finally, we demonstrate that we can reduce the model-size by up to 7x using modern weight quantization methods.